Audit Logging for AI Agents
Audit logging for AI agents means recording all agent actions, decisions, tool calls, and outcomes in a structured, tamper-resistant log. EU AI Act Article 12 requires this for high-risk AI systems to enable accountability, debugging, and regulatory compliance.
Frequently Asked Questions
What should AI agent audit logs include?
Log: (1) All LLM calls with prompts and responses, (2) Tool invocations with parameters and results, (3) Decision points where the agent chose between actions, (4) User inputs and agent outputs, (5) Errors and exceptions, (6) Timestamps and session identifiers.
What does EU AI Act Article 12 require for logging?
Article 12 requires automatic recording of events relevant to identifying situations that may present risks. For AI agents, this means logging all actions that could affect users, including tool calls, decisions, and data access patterns.
How does Inkog check for missing audit logging?
Inkog identifies AuditLogNode absence in agent workflows. It flags tool calls without logging, decision points without recording, and agent actions that aren't traceable — mapping gaps to Article 12 requirements.
How Inkog Detects This
Inkog verifies that agent workflows include audit logging by checking for AuditLogNode presence. It flags tool calls, decision points, and external API calls that lack corresponding log entries, mapping findings to EU AI Act Article 12.
npx -y @inkog-io/cli scan .Related Terms
Check Logging Compliance
Scan your AI agents for vulnerabilities. Free for developers.
Start Free Scan